Enhanced Driver Stress Prediction from Multiple Biosignals via CNN Encoder-Decoder Model.

Area of Science:

• Physiology
• Machine Learning
• Human-Computer Interaction

Background:

• Driver stress negatively impacts road safety.
• Accurate stress detection is crucial for developing effective driver support systems.
• Existing methods often lack robustness in classifying diverse stress types.

Purpose of the Study:

• To introduce PhysioFuseNet, a novel framework for driver stress state classification.
• To evaluate the effectiveness of multimodal biosignal fusion for stress detection.
• To compare the performance of different machine learning classifiers within the PhysioFuseNet framework.

Main Methods:

• Acquisition of multimodal biosignals (ECG, EDA, PPG, respiration) from 25 healthy subjects in a driving simulator.
• Utilizing a CNN-based encoder-decoder model for feature extraction.
• Employing intermediate fusion of extracted features.
• Classification using Support Vector Machine (SVM) and Random Forest (RF) algorithms.

Main Results:

• PhysioFuseNet achieved high classification accuracies: 0.95 for SVM and 0.94 for RF.
• The framework demonstrated superior performance compared to previous methods.
• The system showed particular strength in classifying emotional and cognitive stress states.

Conclusions:

• PhysioFuseNet offers a robust and effective approach for driver stress classification.
• Multimodal biosignal fusion significantly enhances stress detection accuracy.
• The proposed framework holds promise for real-time stress assessment applications in both clinical and real-world driving scenarios.